Search

Due to the excessive temperature rise in the Central Processing Unit (CPU), computers shut down and system damage occurs over time. In this study, a new thermoelectric cooling system is designed to reduce the temperature in the CPU. The novel cooling system is designed using a thermoelectric module. It is to remove the excess heat by conduction and convection by taking advantage of the temperature difference between the thermoelectric cooler and the CPU we add to the system. Since the temperature of the thermoelectric cooler will always be lower than the CPU temperature, effective cooling will be provided. A special electronic circuit and software have been developed for the control of the cooling unit. Three different artificial intelligence models (artificial neural network, random forest, and k-nearest neighbor) were created to dynamically control the additional cooling system and their successes were compared. Artificial intelligence determines the power and fan speed of the thermoelectric cooling system. It performs this control by evaluating all parameters (different values such as CPU frequency, voltage, number of processes) instead of a specific CPU load or a specific temperature value. While the CPU temperature was 41°C at maximum load, this temperature was reduced to 31°C thanks to the designed thermoelectric cooling system. All methods provided a high classification success in education. By using the artificial neural network 97.973% classification success was achieved, whereas the classification success for random forest and k-nearest neighbour model were 97.297% and 96.306%, respectively. [ABSTRACT FROM AUTHOR]

Forest fire is a disaster that destroys all living and non-living beings in the unity of life in the forest by burning and has economic and ecological damages. In recent years, temperatures and droughts that have been above the seasonal norms due to global warming have increased the risk of forest fires. In order to minimize the damage caused by forest fires, early warning, fast and effective intervention is very important in firefighting. Machine learning methods are used in early warning systems today. In this study, it is aimed to control and reduce the effects of fires by predicting possible forest fires in order to fight forest fires. The dataset for the wildfire prediction model was developed from the official website of NASA's Oak Ridge National Laboratory (ORNL) Center for Distributed Active Archives (DAAC). A forest fire prediction model was created by processing these data with machine learning methods. The data set was adapted to the classification model by applying various preprocessing steps. With the feature selection techniques, the least number of feature subsets were selected by providing the highest level of data integrity without using the entire data set. By choosing the most important and useful features in finding the target variable, a model was created with 6 different classification algorithms, namely Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, K-Nearest Neighbor and Naive Bayes. Validation process was performed to evaluate model performance and hyperparameter optimization was performed for best parameter selection. Among the classification algorithms used in this study, an accuracy rate of 97% was obtained with Random Forest and 96% with Naive Bayes, which is one of the two most successful algorithms with the validation process. [ABSTRACT FROM AUTHOR]

With digitalization, the world of crime has also become digital and the number of crimes committed over the internet is increasing day by day. Cybercriminals and attackers use secret networks on the Internet, called the Dark Web, to hide their identities and provide encrypted communication. Darknets have different and special access methods than normal internet infrastructure. All access to these networks is suspect and needs to be investigated. Because the Darknet provides encrypted communication, it is difficult to detect and classify with today's security tools. In this study, only the statistical information of packets was analyzed using machine learning approach without deciphering encrypted network traffic. CICDarknet2020 dataset was used and a detailed experimental study including K Nearest Neighbor, Logistic Regression, Random Forest, SVM, Decision Tree, Gaussian Naive Bayes, Linear Discriminatory Analysis, Gradient Boosting, Extra Tree and XGBoost algorithms was carried out for packet analysis. In experimental studies, it has been observed that the Decision Tree algorithm has the highest classification success with an accuracy rate of 93.32%. [ABSTRACT FROM AUTHOR]

The increasing urbanization and rapid economic development have led to growing housing markets. This has resulted in a rising demand for housing due to migration towards cities and the acceptance of housing as both an investment and a wealth protection tool. On the other hand, housing prices are affected by many factors, thereby studies on housing price forecasts are carried out within the scope of different models and various methods. In line with the significant importance of housing markets as one of Turkey's economic growth tools, this study includes the forecast of the classification of 4-bedroom (3+1) housing prices in the central district of Kütahya province. The dataset used includes various criteria that affect housing prices, and different machine learning classification algorithms that are utilized. As a result, it is seen that the accuracy rates of the k-Nearest Neighbor (k-NN), Decision Trees (DT) and Random Forest (RF) algorithms were above 60%. As the main performance criterion, the AUC scores yielded that the most successful classification forecast is acquired by the RF algorithm, which is followed by the k-NN. In the context of the variables and data set included in this study, it was seen that successful outcomes in the classification forecasts of 4-bedroom apartments’ housing prices in the city center of Kütahya were obtained by these two methods. [ABSTRACT FROM AUTHOR]

Heart diseases are diseases with a high mortality rate. Clinical methods and machine learning methods have been used frequently in the diagnosis of the disease. In this study, the KNN algorithm was used for diagnosis. In order to increase the performance of the algorithm, parameter adjustment has been made and in this context; Manhattan, Euclidean and Chebyshev distance measurements, Uniform and Distance weighting methods and neighbor numbers between 1...15 have been tested on UCI Statlog and Cleveland datasets. The highest classification accuracy for the Statlog dataset is 67.90%, which is obtained with the number of neighbors = 5, the distance method = Euclidian, and the weighting = Distance. Genetic algorithms were also run on the same data set and 88.88% accuracy was obtained for the number of neighbors = 5, distance method = Euclidean and weighting = Distance. While the maximum classification accuracy obtained for the Cleveland dataset was 71.42% before optimization, it was measured as 90.11% after optimization. The parameters that give the highest classification accuracy for the Cleveland dataset are; number of neighbors = 3, distance method = Manhattan and weighting = Uniform. [ABSTRACT FROM AUTHOR]

It is thought that data mining can solve many problems in visible light communication. In the VLC systems, different receiver and transmitter units are usually designed for each modulation technique. Here, the proposed systems are designed for shortened pulse position modulation. In the paper, K-Nearest Neighbor algorithm and Decision Trees based demodulators that can be applied to synchronous modulation techniques are proposed. Afterwards, a generalized entropy expression is obtained for M-SPPM signals. Additionally, demodulators that can decode signals according to the Manhattan and Minkowski distance and other demodulators are compared in terms of accuracy and bit error rate. From the numerical results, it can be seen that the proposed KNN and Decision Trees based demodulators can be used as traditional systems for synchronous modulation techniques. Furthermore, it is seen that bit error rate performances and accuracy rates of the proposed KNN based demodulator and decision trees based demodulator that is optimally designed are completely the same. In 4-SPPM signals, it is seen that the accuracy of KNN (Demodulator-1) and Decision Trees (Demodulator-3) is 98.78% for SNR=10 dB. Here, the accuracy of the KNN (Demodulator-2) and Decision Trees (Demodulator-5) is 99.07% for 8-SPPM signals. [ABSTRACT FROM AUTHOR]

The widespread use of the internet today has increased the importance of internet security. The compromise of personal information, passwords, and other sensitive information by hackers or fraudulent websites creates serious risks for internet users. To use the internet securely, users need to be aware and take precautions for their security. For example, they should change their passwords frequently, use secure connections, and protect electronic devices (such as computers, phones, tablets, etc.) with up-to-date security software. In addition, protecting the devices we use with Firewall technology is of great importance for security. In this study, a classification process was applied using the NN (Neural Network) and K-NN (K-Nearest Neighbor) algorithms with 65,532 log records obtained from Firat University's firewall device. Moreover, feature selection techniques were used to determine the importance and similarity rates of the columns in the dataset. In the Neural Network algorithm, the 'Adam' function optimization resulted in 98.46% success, while in the K-NN algorithm, the most successful result was achieved with a k-value of 20 and in the Manhattan distance with a success rate of 99.08%.In a previous study using the same dataset, the most successful technique among four SVM techniques was achieved with a success rate of 98.5% using the Sigmoid function. [ABSTRACT FROM AUTHOR]

The traditional karyotype studies are widely used in plant systematics to evaluate the positions of species. However, studies sometimes can not solve systematic problems. In recent years, computer-based systems have gained importance in contributing to the solution of the taxonomical problems. The aim of this study was to identify three Lallemantia species by using the LBP (local binary pattern) texture operator based on chromosome images in mitosis. LBP is the one of the most powerful and easily applicable tool for identifying local image patterns. In this study, microphotographs of 641 cells in the metaphase stage of mitosis were used. The LBP involves preprocessing, feature extraction, feature selection, and classification. Decision tree (DT), linear discriminant (LD), support vector machine (SVM), K-nearest neighbor (KNN), bagged tree (BT) and ensemble subspace nearest neighbor (SNN) were used in the classification stage. This study found that the best acting classifier was SNN because achievement rate was one hundred percent. Also, a dendrogram was formed to measure the similarity among the three species. As a result, LBP can be accepted as a tool for classifying plants by using chromosome images. [ABSTRACT FROM AUTHOR]

Since December 2019, the Covid-19 virus afftected our lives and continues to affect the whole world significantly. The investigistion of the indicators of the Covid-19 virus and vaccination studies are of great interest to overcome the Covid- 19 pandemic based on the World health organization recommendations. In this context, many scientific studies have revealed valuable information for the future of the virus. In this study, estimation of the cOvid-19 cases and Classification of changes in the daily number of cases in Turkey was carried out by using support vector machine and k-nearest neighbor algorithms. The indicators that play a critical role in the estimation of the daily patient number classification have been determined as "positivity rate", "fillation rate", "workplace mobility" and "mobility in parks". It has been observed that the k-nearest neighbor algorithm (84.7%) is the most successful algorithm in the estimation of the daily number of cases when considering the highlighted features. [ABSTRACT FROM AUTHOR]

Objective: In order to diagnose myeloproliferative diseases, the presence of genetic mutations is examined by the clinician in line with the disease diagnosis scheme determined by the World Health Organization. In this study, it is aimed to predict the appropriate diagnostic screening test with the model created based on bone marrow and complete blood count findings of patients who applied to the clinic. Accordingly, it is aimed to save time and financial for patients who apply to the clinic. Material and Methods: Prediction was made using the machine learning algorithms by considering the findings of patients referred to the Başkent University Ankara Hospital, Department of Medical Genetics Genetic Diseases Diagnosis Center. Descriptive statistics of the study data were given as median, and statistically significant differences were investigated using the Kruskal-Wallis test statistic. Classification algorithms such as Naive Bayes, K-Nearest Neighbor, Linear Discriminant Analysis, Support Vector Machines, Entropy Based Classification and Decision Tree were used in the estimation. With the algorithms, determinative values such as accuracy, specificity and sensitivity were obtained. The estimations made were examined according to the accuracy rates obtained and the best model was tried to be selected. Results: Statistically significant differences were found between the examined complete blood count values and the presence of mutation. The accuracy rates were found to be around 60% in the models created with Naive Bayes, K-Nearest Neighbor, Linear Discriminant Analysis, Support Vector Machines, Entropy Based Classification and Decision Tree Algorithms. Conclusion: Although the accuracy rate obtained from the machine learning algorithms used is at a moderate level, it was concluded that the results of the study would make a significant contribution to the field, since similar studies have not been included in the literature. [ABSTRACT FROM AUTHOR]

Objective: Breast cancer; it occurs in the tissue cells of the milk ducts in the breast tissue. The uncontrolled increase in the cells forming the milk ducts is called ductal hyperplasia. It is known that the risk of developing invasive (with a tendency to spread) breast cancer in a woman during her lifetime is 13.3%, and the risk of developing breast cancer increases with age. The Gail model is a well accepted cancer risk assessment model which evaluates the main factors in breast cancer. The aim of this study is to compare machine learning methods in breast cancer risk assessment based on the Gail model. Material and Methods: Firstly, the risk factor was determined by the application of the Gail model into the data set, discrete training test data sets were presented which is 80% train and 20% test. Afterwards, k-nearest neighbor, artificial neural network (ANN), support vector machine (SVM) and naive Bayes (NB) algorithms applied on each set and risk estimation results were compared. Results: Classification performance from the lowest to the highest for 80% training and 20% test data set according to the comparison results is as follows; SVM [area under the curve (AUC)=0.911], NB (AUC=0.939) and ANN (AUC=0.949). Conclusion: Early diagnosis of breast cancer increases the number of possible treatments, the success rate of the treatments and the chance of survival. It has been seen that machine learning algorithms effective in breast cancer risk calculation. [ABSTRACT FROM AUTHOR]

This study aimed to predict the 1 to 2 year future time of the financial failure of 86 manufacturing companies that are operating in Borsa İstanbul. The data comprised of 2010-2012 period, and it depends on 8 quantitative financial variables. Beside 6 variables come from non financial statements. In the study, Artificial Neural Network (NN), Classification and Regression Trees (CART), Support Vector Machine (SVM) and k-Nearest Neighbors (KNN) were used to compare classification performances of related methods. ROC Curve was used to compare the classification performance of the methods. As a result of the analyseis, the overall classification accuracy from the highest to the lowest was SVM (92,31%), CART (88,46%), ANN (84,62%) and KNN (80,77%) 2 years before the financial failure. The overall classification accuracy from the highest to the lowest was CART (96,15%), ANN (92,31%), SVM (80,77%) and KNN (84,62%) 1 year before the financial failure. Return on Equity (ROE) and Return on Assets Ratio (ROA) were found as important variables in the creation of the CART decision tree. The fact that the four models obtained in thise study predicted financial success/failure at a higher rate, and it shows that the models obtained in this study can be included in the models used by relevant people. [ABSTRACT FROM AUTHOR]

In our coun try, the number of small ruminant animals is decreasing day by day due to various reasons. In parallel with the decrease in the number of small ruminants, significant decreases are seen in animal production. One way to prevent the reduction in the number of small ruminants is to be able to make successful predictions and analysis related to the diagnosis. Thanks to computer-aided diagnostic studies performed with machine learning, the quality of health services increases while the costs of the health sector decrease. The aim of this study is to perform computer aided diagnosis in neonatal lambs using machine learning methods. Hence in study, decision tree, naive bayes, k-nearest neighbors, artificial neural networks and random forest methods were used. The performances of these classification methods were analyzed with accuracy, balanced accuracy, specifity, recall, F-measure, kappa and area under the ROC curve (A UC) criteria. As a result of the study, the Naive bayes method more successful results than other methods for computer aided diagnosis produced. It is very important tha t, the Naive bayes method is simple and easy to apply, achieves more successful results than other complex methods. [ABSTRACT FROM AUTHOR]

The coast of Eastern Mediterranean located on downstream of the Göksu River has been exposed to flood disaster for several times in the past. Many studies in the basin have been carried out to prevent the developing damage and use advantage of the river. One of them is the Kayraktepe Dam which has been on the planning stage for more than thirty years. In this study, the information about historical process of the Kayraktepe Dam will be given and evaluation the flood frequency curves with nonparametric approach based on K nearest neighbors (KNN) in order to serve one of the purposes of the dam will be emphasized. While these results are compared with aforementioned parametric approaches as well as the results of Electrical Power Resources Survey and Development Administration which was responsible for performing initial investigation and planning of Göksu Basin, the applicability of the nonparametric approach will be questioned. Although quantile estimates for 500 years return period was 2363,95 m³/s in past feasibility report, this value descended to 1814,83 m3/s with nonparametric approach. The decreasing of designing dam dimensions is important to reduction both economical and environmental concern. As a result, the study will offer a detailed and a wide range of discussion on the design discharge of the dam. [ABSTRACT FROM AUTHOR]

The combinations of bases in a DNA sequence correspond to a gene in that DNA sequence, RNA copy sequences are extracted from these genes. When these copied RNA's extracted, the base sequence of gene is not read from the beginning to the end completely. The uncoded and unreadable section of the gene is called 'intron' and the coded section of the gene is called 'exon'. Where is a protein coded? How much is encoded? Where are growth and development regulated? Where are stem cells converted to other cells? The answer to all of these questions and the investigation of genetic diseases, such as cancer, are possible by DNA sequences that can be classified as the exon and intron. The aim of this study is to compare the performance of different digital mapping techniques for classification of DNA sequence as the exon or intron. For this purpose, DNA sequences of the MEFV gene in human species are transformed to numeric sequences by nine different digital mapping techniques. The Discrete Fourier Transform Method (DFT) is used to classify these transformed sequences. Four different windowing functions are used and their classification performance are compared in this method. Also, the results obtained from the Fourier-based method have been compared using the Support Vector Machine and the K-Nearest Neighbor methods. Integer mapping technique achieved the highest classification performance with 96.2% in the DFT method than other machine learning methods. Classification performance in Hamming windowing function is higher than other windowing functions. [ABSTRACT FROM AUTHOR]

As one of the ensemble methods, Mixture of Experts is used to gain higher prediction performance in classification area. In this technique, a dataset is divided into regions by soft clustering. An expert for each region is assigned and trained with the samples of the corresponding region. In this study, a dataset is divided into regions by a hard clustering method and a decision tree classifier is constructed for each region as an expert. Each expert has their individual decision for a test point. The class prediction for the test point is performed by the proposed gate function which aggregates all the decisions of experts. In calculations of the final decision via the gate function, each expert has different weights according to its distance to the test point. In the experiments on a group of benchmark datasets, better performance has been obtained with the new gate function. This new method gives better results than base classifiers such as decision trees and k Nearest Neighbors. [ABSTRACT FROM AUTHOR]

In closed areas, fingerprint location estimation algorithms using a radio map, which can be used in wireless sensor networks, consist of two phases, mapping and location estimation. There are several important criterions in the phases, for example using of a radio map which has a small capacity and which shows a comprehensive RSSI dispersion, a quick position calculation, a good accuracy rate, and so on. In the study, a novel technique which employs K-Means method to decrease size of the radio map by reducing separately RSSI (Received Signal Strength Indicator) data of the anchors has been proposed for indoor position detection in WSNs. Besides, a subfield construction process to increase the accuracy of the estimation has been carried out. In the location estimation phase, a technique which is different from K-Nearest Neighbour (KNN) has been preferred. In this technique, unlike KNN the number of the decision cells varies dynamically according to RSSI data received. The system was implemented in a closed environment by using TELOSB nodes. The results of the experiments and the calculations were compared with the results of well-known deterministic methods based on KNN and the validation of the proposed system was tested. [ABSTRACT FROM AUTHOR]